Control problems in manufacturing and communication networks

制造和通信网络等复杂互联系统的控制提出了重要的概念和计算挑战。近似系统分解是部分减轻这些挑战的一种方法。研究人员及其同事最近开发了一种用于制造转移线的分解方法，该方法允许新的结构见解，并大大简化了重要的相关最佳缓冲尺寸问题。我们建议将这种分解方法扩展到两个重要的方向：首先，生产控制体系结构比看板（每台传输线机器都有一个固定的缓冲区，它必须尝试填充）更通用，例如CONWIP，即恒定在制品（看板策略应用于一串机器），或广义看板（从生产授权的迁移中去同步实际生产）；第二，多部件（相对于单个部件）类型系统，因为它们构成了大部分资本密集型柔性制造系统。解决这些制造系统问题的专有技术将被转移，以帮助解决下一代通信网络（Internet和无线系统相互连接以提供具有性能保证的类似Internet的服务）中相关的服务质量（QoS）保证问题。事实上，固定的互联网路由与传传线非常相似，而所谓的DIFFSERV范式（将互联网流量聚合到少量不同的QoS类中以实现可扩展性）中的性能分析与制造业中多部分类型传传线的性能分析密切相关。最后，已经观察到流体近似可以在分析离散制造过程中产生有趣的见解和可接受的数值精度。利用无线数据传输过程的流体近似，我们建议进一步探索已建立的（连续）线性调节器理论在推导具有公平和效率正确组合的可处理调度方面的潜力。

The control of complex interconnected systems such as manufacturing and communication networks presents important conceptual and computational challenges. Approximate system decomposition is one way of partially mitigating these challenges. The researcher and coworkers, have recently developed  a decomposition method for  transfer lines in manufacturing  which permits new structural insights and greatly simplifies the important associated optimal buffer sizing problems. It is proposed to extend this decomposition method in two important directions by considering:   First, production control architectures more general than Kanbans ( for which each transfer line machine is associated with a fixed buffer that it must attempt to fill), such as CONWIP, i.e. CONstant Work In Process (the kanban strategy is applied to a string of machines), or generalized kanbans (desynchronizing actual production from migration of production authorization); second, multi-part (versus single part) type systems, as they constitute the bulk of capital intensive flexible manufacturing systems. The know how developed in solving these manufacturing system problems is to be transferred in order to help address related quality of service (QoS) assurance issues in next generation communication networks (Internet and wireless systems interconnected to offer Internet like services with performance guarantees). Indeed, fixed Internet routes strongly resemble transfer lines, and performance analysis  within the so-called DIFFSERV paradigm (aggregating Internet traffics into a small number of differentiated QoS classes for scalability) is closely related to performance analysis of  multi-part type transfer lines in manufacturing. Finally, it has been observed that fluid approximations can yield interesting insights and acceptable numerical accuracy in analyzing discrete manufacturing processes. Using fluid approximations of wireless data transmission processes, we propose to further explore the potential of the well established (continuous) linear regulator theory in deriving tractable schedules with the right mix of fairness and efficiency.